Simulation and Assessment of Vertical Scaling for a Smart Campus Environment Using the Internet of Things

The concept of scaling in the Internet of things (IoT) refers to the capacity of expanding the number of Internet-connected devices. From the perspective of exponentially increasing IoT devices, scaling is an important research topic. According to a future flexibility point of view, a smart campus holds significant importance and requires an in-depth study. This study presents a new scalability categorization, comprising the device layer, gateway layer, communication layer, and server/cloud layer. Furthermore, the transport system of the smart campus is assessed and analyzed at the server layer using a custom-based simulator created in Visual Studio. According to the outcomes, raising the workload causes the server’s response time to increase. Response time is reduced as a result of scaling up. When scaling up to a specific point and raising the workload, response time further increases resulting in demanding horizontal scaling in the future. This study is expected to aid in determining the capabilities of current and upcoming smart transport systems in the context of smart campuses.

IoT is a system of networked objects such as sensors 68 (smartphones, vehicles, buildings, etc.) and actuators that 69 generate a significant amount of data, which results in big 70 Data. The sensing devices continually gather and transfer 71 data to the IoT server. Big data is termed as 4Vs: volume, 72 value, velocity, and variety. In the decade 2010-2020, the 73 data capacity rose forty-four times and reached 35 ZB from 74 0.8 ZB. As a result, future IoT applications may face scal-75 ing difficulties due to the massive influx of data from IoT 76 objects [10]. Scaling, in the IoT, refers to the future expan-77 sion of linked objects with the Internet which are used for 78 detecting, updating, monitoring, and sending data in order to 79 create information and gain understanding to accommodate 80 more data processes over the normal case. 81 The discussed research works are analogous to such 82 research highlights in terms of framework, a number of gad-83 gets, gateways/routers, transmission technologies, and server 84 types; however, the layout of the selective campus, the num-85 ber of gateways, devices, transmission technologies, and the 86 kind of server machines are all dissimilar. The created sys-87 tem uses instantaneous methods and transmits available data 88 to make decisions, which is the research's main innovation. 89 Keeping these points in view, this study participates toward 90 scalability solutions and makes the following key contribu-91 tions 92 • A new scaling classification approach is presented 93 which is divided into four layers: device layer, gateway 94 layer, communication layer, and server layer. The clas-95 sification is based on IoT infrastructure. Gateways are 96 used at the edge to provide wired or wireless connectiv-97 ity to IoT objects. Devices are employed so that verti-98 cal and horizontal scaling may be implemented, much 99 like at gateways. The cloud/server is connected to the 100 gateways via wired or wireless connection. As a result, 101 scalability may be used at both the communication and 102 server levels.

103
• The vertical scaling is performed and assessed of the 104 transport system for the smart campus at the server layer. 105 A custom-built simulator is utilized, as a case study, for 106 accessing the vertical scaling of a smart transport system 107 at the server layer.

108
• Performance is analyzed regarding different parameters 109 like the number of buses, response time, etc. Perfor-110 mance with existing studies is also carried out to analyze 111 the efficiency of the proposed approach.

112
The remainder of the article is explained as follows. 113 Related work is presented in Section II, Scaling is explored 114 in-depth in Section III. Section IV discusses scaling testing. 115 Results and discussion are presented in Section V, and the 116 conclusion is ultimately finalized in Section VI.

118
The number of research studies has boomed for IoT during 119 recent years and several important aspects of IoT technol-120 ogy have been investigated. For example, a container tech-121 nology is suggested by Aruna and Pradeep in [11] to link 122 many IoT objects at gateways for IoT applications. The tech-123 nology is utilized to store and process data, allowing the 124 network to grow while improving network proficiency. The 125 container technique is utilized in clusters in a distributed fash-126 ion. In [12], the authors suggested a decentralized method for 127 scalability in which a hybrid cloud computing environment 128 (private for delicate and public for large data) is used for the 129 management of data to exchange data among many partners.
regions, accumulating the gathered data, and communicating 144 to decrease power consumption.  intelligence algorithm is used to create and implement a data 186 analysis and behavior prediction system for computing GPS 187 user trajectory.

188
After identifying the required criteria of a smart uni-189 versity/campus, Fernández-Caramés and Fraga-Lamas [18] 190 explored state-of-the-art transmission techniques and 191 blockchain-type architecture. In [19],  cussed the scaling, mobility, and availability of IoT improve-193 ments in health care, as well as portability and privacy issues 194 of IoT. In [20], the authors looked into the security trade-offs 195 of the smart home. 196 Nie [21] introduced the use of cloud computing and the 197 IoT in education. The discussion of the current state of smart 198 campuses was followed by an explanation of how digital 199 campuses differ from smart campuses in terms of both old 200 and new technologies. By developing a model and application 201 structure for a smart campus based on cloud computing and 202 the IoT, examining how to use it, and lastly talking about how 203 to implement it extensively.

204
Cloud computing and associated technologies were pre-205 sented by Li [22], who then examined the requirements for a 206 smart campus service platform and, in light of those findings, 207 created a smart campus service platform based on cloud com-208 puting technology to support the growth of smart campuses. 209 Innovative tracking methods in automobile or transporta-210 tion systems, among other IoT-based applications, demand 211 the movement of the IoT device across various IoT technolo-212 gies. Ayoub  In an IoT system, scaling also discusses the upcoming flexi-224 bility. In case the setup does not fulfill varying needs, it is nec-225 essary to restructure it, which is budget and time-intensive. 226 Vertical and horizontal scaling are the two forms of scaling 227 that may be distinguished with respect to flexibility [24]. The 228 next sections go through the specifications of each kind. Vertical scaling is also termed scaling up. Vertical scalability 231 implies the capacity of an IoT server, transmission media 232 bandwidth, gateways, and devices to facilitate additional data 233 than typical function [25]. It belongs to the capability and effi-234 cacy of assets such as executing capability, communication 235 bandwidth, and storage ability of linked devices in fulfilling 236 varying desires. The advantages of this form of scaling are 237 that it involves less energy than many nodes, it is simpler 238 to control owing to a single machine unit, and cost, soft-239 ware implementation, arrangement, installation, and all are 240

283
Normally, the communication channel's uplink capacity is 284 relatively lower [13]. IoT sensors collect data from the envi-285 ronments and transmit it to the IoT server/cloud through an 286 uplink bandwidth. Vertical scaling is applied as the transmis-287 sion capability of a communication link is raised. Horizon-288 tal scaling is applied as the number of transmission links is 289 expanded.

291
Vertical scaling is applied as the processing capability, stor-292 age capability, and hard disk space of a server at the 293 server/cloud layer are raised, while horizontal scalability is 294 used when the number of servers is increased.

297
The transport system of a smart varsity campus is nominated 298 as a research study in the modern age of IoT when every field 299 of life is attempting to be smart by utilizing IoT technologies. 300 A bespoke virtual data source/sender and server/sink (Emu-301 lator) is created to evaluate the scaling ability. Utilizing the 302 Microsoft Transport datasets [28], [29], [30], a smart trans-303 port system for a varsity campus is simulated in this study. 304 The dataset is in a comma-separated values (CSV) format 305 and comprises 17621 records. Each record contains paths and 306 contains geographical information including latitude, longi-307 tude, and altitude. Moreover, it has global positioning sys-308 tem (GPS) information, as well as, the date and time stamp 309 collected at different time spans over four years. Using this 310 information, Microsoft Visual Studio using C# (C-Sharp) is 311 used to develop a customized simulating software with virtual 312 data senders and a server.

313
Before designing the simulator, different classifiers are 314 applied to the dataset to evaluate the performance of these 315 classifiers. For this purpose, k nearest neighbor, Naïve Bayes, 316 logistic regression, decision tree, and support vector machine 317 is utilized. Table 2 presents the accuracy, precision, and recall 318 of the above-mentioned classifiers. The statistics show that 319 the decision tree outperforms the selected classifiers. 320 VOLUME 10, 2022 The comparison of classifiers is also depicted in Figure 4. where n is the number of GPS data packets executed in a 355 particular executing cycle.

356
The cycle delay uplift with the expansion of b buses as a 357 result of greater data executing costs; the complete execution 358 workload is expressed as under where b is the counting of buses whose GPS data is delivered 361 to the IoT receiver/server virtually.

362
The server's average latency is now computed and written 363 as follows The aggregate GPS data packets in the queuing buffer at 366 the last of the execution cycle are represented by N .

367
On normal load circumstances, Figure 5 represents a graph 368 of data execution time in ms at several time spans. It is found 369 that execution time varies with time and peaks after a specific 370 duration. When the load on the virtual IoT server is raised 371 by adding additional buses, the time it takes to complete the 372 task increases, resulting in a rising curve in Figure 6. Simi-373 larly, scaling up the IoT server diminishes processing time by 374 reducing the least element of single GPS data packet delay, 375 for instance, from 10 to 2 with five steps by decreasing 2 376 each time. In the random function, the decreasing fixed value 377 widens the lowest and largest range for individual GPS data 378 packet delay, effectively increasing the chance of minimal 379 number choice. Figure 7 shows a falling curve.

380
Additionally, packet loss is another significant issue that 381 degrades server efficiency. It happens when the IoT server 382 is overloaded with packet processing and gains more time 383 than the virtual IoT sender delivering GPS location packets. 384 For instance, the queue will begin to enlarge, and its size 385 will continue to rise till it extends to the highest capacity of 386 100 data packets. On this occasion, the IoT server will discard 387 the virtual senders' received packets, causing them to drop the 388 packets.

390
The following four phases make up the primary algorithm 391 for assessing vertical scalability. The algorithm's flowchart 392 is presented in Figure 8.   big. When a light workload is imposed, the simulation 417 effortlessly executes the GPS packets. Figure 5 represents a 418 graph presenting standard operation, with the x-axis showing 419 system time and y-axis representing processing time. When 420 an average workload is introduced by expanding the number 421 of buses, the simulation performs normally and gives a rea-422 sonable behavior of how the setup works. As soon as there 423 is surging in packets caused by an increase in the number 424 of busses in certain rush conditions, for example, admission 425 process days or exam days, or just by an increase in the 426 number of enrolled students, the simulator displayed a graph 427 overburdening, as illustrated in Figure 6.

428
As seen in Figure 7, when scaling up or vertically scaling 429 is utilized, the average execution time reduces again. As the 430 number of enrolled students is rising, and smart varsity costs 431 are also rising, as a result, the smart campus will expand 432 admitting more students so the number of busses will also 433 VOLUME 10, 2022   Table 4 compares the assessment of smart campus trans-448 portation to prior studies. The load balancing of smart trans-449 portation utilizing vertical scaling assessment is explored, 450 which is a completely new feature of the smart campus. Prior 451 to this study, only the smart campus or smart transportation 452 were studied separately, and in some cases, just architecture 453 was offered, as in [22]. The algorithms are proposed along 454 with statistical analyses, such as accuracy, precision, and 455 recall, due to the more technical dive into the scenario. 456 Normally, in previous work, the debate was theoretical 457 rather than technical. Because algorithms are provided with 458 statistics, the work is completely new and unique in this 459 regard. Table 4 displays entries marked as ND indicating that 460 these parameters are not discussed. Several aspects are not 461 covered by the existing literature, as shown in Table 4, like 462 selected method, datasets, the context of the study, etc.

493
The unique taxonomy explains scalability types in the case 494 of vertical and horizontal scaling (IoT future flexibility) and 495 device layer, communication layer, gateway layer, and server 496 layer (IoT infrastructure). A custom-built simulator is uti-497 lized to test vertical scaling at the server layer. There are 498 three sorts of loads: regular, medium, and large. The average 499 execution time of the IoT server raises when the workload 500 is high. After that vertical scalability is used until a specific 501 point is reached, at which point the average execution time 502 is reduced. When the workload on the virtual server is raised, 503 the average execution time of the virtual IoT server also rises. 504 The findings demonstrate that when the workload on the IoT 505 server is high, the execution time of the IoT cloud server 506 increases and becomes a bottleneck, necessitating horizon-507 tal scaling with many IoT cloud servers in the future. This 508 study opens possibilities of how to use scaling while creat-509 ing IoT systems at various layers, with appropriate vertical 510 or horizontal types, to make the system adaptable for the 511 future.